A welfare trap? The duration and dynamics of social assistance use among lone mothers in Canada.
Cooke, Martin
BETWEEN THE LATE 1980S AND THE early 2000s, Canada and other
western welfare states engaged in important and contentious debates
surrounding the restructuring of welfare state programs. These occurred
partly as a response to relatively high rates of benefit use and rising
government debts resulting from recessions and economic restructuring
over the period. The policy responses have involved the redesign of
programs to further encourage work and to reduce "dependency"
on benefits and to increase labor force participation, and have occurred
in continental European and Scandinavian welfare states, as well as in
"residual" or "liberal" welfare states, such as
Canada and the United States (Esping-Andersen 1999). The thesis of the
"path dependency" of welfare state restructuring (Pierson
1994) suggests that changes in these countries have generally been in
keeping with the core characteristics of their regime types. Whereas the
European Commission's Flexicurity project represents current
efforts to balance traditional social protection with labor market
flexibility (Wilthagen 2007), changes in the Anglo-American countries
have followed the principle of providing low levels of support so as not
to create disincentives for labor market participation (Myles and
Pierson 1997; Pierson 1994).
In Canada and the United States, the changes from the late 1980s to
the early 2000s have generally included reduced eligibility for welfare
or social assistance, the lowest tier of state income security, combined
with increasing requirements for work-related activities as conditions
of assistance, in order to move welfare recipients into the labor market
(Bashevkin 2002). These changes were facilitated in Canada by the 1996
shift in federal-provincial funding arrangements from the shared-cost
Canada Assistance Plan (CAP) to the fixed-amount Canada Health and
Social Transfer (CHST). This change significantly reduced the funding
available for provincial welfare programs, while also removing the
CAP's requirements that social assistance be provided on the basis
of need alone. At roughly the same time, the Canadian government made
important changes to federal Employment Insurance (EI), including
reducing eligibility and the length of benefits, forcing some unemployed
workers to turn to provincial social assistance.
The CHST and the changes to EI, it is argued, encouraged the
provinces to focus on reducing welfare expenditures by changing benefits
and eligibility requirements (Bashevkin 2002; Boychuk 2006; Campeau
2005). The most dramatic of the Canadian changes occurred in Ontario,
where benefit levels were reduced by 21.6 percent in 1996 and where a
new requirement of mandatory work-for-welfare, or "workfare"
was imposed. Other provinces also made social assistance benefits
conditional on participation in job searches or training, or taking the
first available job (Gorlick and Brethour 1998).
The impacts of these changes on the Canadian social assistance
caseload are somewhat difficult to identify because they coincided with
a period of economic and employment growth. The changes do seem to have
corresponded with a reduction in the rates of social assistance use, but
may also have contributed to a rising intensity or "depth" of
poverty in Canada over the 1990s (Picot et al. 2003). Although a
majority of those who left social assistance over the 1990s saw their
economic conditions improve, it appears that for a sizeable minority,
perhaps as large as 30 percent, economic well-being was lower after
leaving welfare (Frenette and Picot 2003).
The changes to provincial welfare systems highlight a number of
assumptions about the dynamics of social assistance use in Canada, and
which are common to other liberal-democratic welfare states. For one,
the policy focus on reducing "dependency" and on encouraging
work effort would seem to indicate that welfare is generally a long-term
phenomenon and that recipients have a tendency to remain on benefits
unless required or forced to leave, or until they have acquired enough
"human capital" or saleable skills that allow them to find
well-paid work (Bane and Ellwood 1994). This is certainly the assumption
of econometric models of the "work or welfare decision";
people rationally weigh welfare benefits against the expected utility of
wage work (e.g., Charette and Meng 1994). It also highlights the
assumption that paid work is the major route off of welfare. This is
particularly important considering that, as in the United States, a
large proportion of Canadian social assistance recipients are lone
mothers with young children. An examination of U.S. Aid to Families with
Dependent Children (AFDC) receipt in the 1980s found that nearly 30
percent of women left welfare through marriage, rather than by finding
work (Bane and Ellwood 1994). Some have suggested that the changes made
under the U.S. Personal Responsibility and Work Reconciliation Act
(1996) were intended to promote exits through marriage, or
"wedfare" as much as to promote work (Mink 2001).
In the Canadian context, there has not been the same concern that
welfare policies might promote marriage as an exit from assistance
receipt, but it has been pointed out how the changes in welfare have
disproportionately affected women. Provinces had generally excluded
women living alone with children from having to participate in paid
work-related activities. Indeed, as social assistance had developed from
mothers' allowances programs after the First World War, the
expectation remained that women's main duty was to raise children,
and that the state would take the role of the absent or deceased husband
(Little 1998). But whereas paid work by lone mothers had previously been
discouraged, provinces increasingly made women, including those with
young children, subject to the same work-related requirements as men
over the 1980s and 1990s. In the absence of flexible employment
arrangements and available and affordable child care, and given the
large proportion of social assistance recipients in Canada who are lone
mothers, changes to social assistance programs are therefore gendered in
their effects, even though the programs are formally universal (McMullin
et al. 2002).
These concerns led us to ask several questions about the use of
social assistance and particularly by lone mothers in Canada. These
include questions about the duration of assistance and the factors that
affect it and the types of events that precipitate social assistance
beginnings and hasten endings. We are particularly interested in the
question of "dependency," and whether social assistance is
generally a long-term phenomenon. As well, we investigate how time on
social assistance may itself affect the chances of leaving it, and
whether there is evidence for the "welfare trap." We frame
these questions in terms of the life course, which helps to think of
social assistance as a temporally limited state that is affected by
previous events and that also forms the context for later events and
decisions.
A LIFE COURSE PERSPECTIVE ON SOCIAL ASSISTANCE AND LONE MOTHERHOOD
The life course, currently popular in a number of disciplines, does
not constitute a theory itself, but rather a framework into which a
number of theoretical perspectives can be incorporated. Typically, the
life course is seen as the pattern of age-graded transitions through
experiences and states in various domains, including the family, work
and education, and health (Settersten 2003). These patterns exist at the
aggregate level, as institutionalized sets of patterns and expectations,
as well as at the level of the trajectories and careers of individuals.
At the institutional level, this is seen in the classic
"tripartite" life course of men, progressing through
education, work, and retirement, and which is largely due to the
institutionalization of public education and retirement systems (Kohli
1986). At the level of the individual, the life course exists as a
distinctive sequence and timing of transitions through states, such as
singlehood, childbearing and marriage, employment, unemployment, and
retirement; and wellness and ill-health.
The theoretical utility of the life course for studying social
assistance lies in its ability to incorporate both economic and
sociological views on the use of these programs. As described by one of
its foremost American proponents, the course of our lives is shaped by
social structures, including gender, race, and social class; by
institutions, and by events in the lives of others to whom ours are
linked (Elder 1994). It is also shaped critically by historical time and
place and by cohort patterns, as well as by developmental processes of
biological and psychological aging. However, as much as the life course
is the product of external influences, this is true in part because
those influences form the context of individual decisions. People make
choices and act, or fail to do so, in the service of a wide variety of
motivations. Decisions and actions taken previously are also part of the
context for subsequent decisions and actions, and affect the various
types of social, material, and physical resources available to address
crises or opportunities (Elder 1994). This is partly captured by the
life-course concept of "cumulative advantage or disadvantage"
(O'Rand 1996). Through such mechanisms as the accumulation or
depletion of human, financial, or social capitals, as well as the
cumulative aspect of physical health, the experience and timing of
previous events can lead to increasingly "positive" or
"negative" trajectories.
Of course, many of these insights are not new, and they can be
related to large bodies of work in developmental psychology, branches of
sociology, economics, social history, gerontology, and health sciences
(Marshall and Mueller 2003). They do, however, provide a useful and
somewhat refreshed perspective on state programs and policies, and a way
of framing important empirical questions related to the effects of
previous life-course trajectories on program use, as well as the
importance of programs for subsequent life events. Welfare programs can
be thought of as targeted to those whose life courses have in some way
deviated from the currently institutionalized pattern. This includes
people out of work for a prolonged period as well as women raising
children alone, and who have either experienced life course events"
out of sequence" by having children before being married, or whose
partnerships have ended (Daly and Rake 2003). The strength of the
incentives to work or marry in these programs reflect the degree to
which they aim to influence the life course by encouraging people to
return to the normative institutionalized pattern. Feminist critics of
welfare state programs have pointed out the significant gender content
in the expected patterns, as well as how that content has recently
changed with the treatment of women primarily as workers or potential
workers, rather than mothers (Little 1998; Mink 2001). From these
perspectives, welfare state programs are part of the reproduction of
gender inequality, reinforcing a male life course that is structured
mainly by labor market transitions and a female life course that is
mainly given its shape by family transitions. Similarly, critiques that
point out the role of welfare programs in maintaining class inequality,
and particularly the way that the highly stigmatized programs discipline
the low-wage labor force (Piven and Cloward 1993), can also be
incorporated into a life-course framework.
As much as this framework can incorporate institutional and
structural influences on the life course, it is also able to consider
individuals' decisions to receive social assistance, or not to, as
part of a strategy by which they try to meet certain goals. This
emphasis on agency and the active construction of one's life allows
it to incorporate economic theories and econometric models that predict
welfare use as a function of individual human capital and expected
wages, the various constraints on working, such as the presence of
children, and the benefits available from welfare. For some
sociologists, these models may not adequately consider the content of
these goals, such as "independence," the importance of paid
work for self-esteem, and how these are socially produced and defined.
However, econometric approaches are useful insofar as they acknowledge
that benefit use is a decision, albeit one made in a highly constrained
context. Similarly, some qualitative sociological research has found it
is useful to consider welfare as one part of a complex strategy by which
lone mothers attempt to provide for their families, both economically
and emotionally (Edin and Lein 1996).
Last, a life-course framework can be used to incorporate
macro-theoretical views on contemporary social change. For example,
theorists, such as Ulrich Beck and his colleagues (1994) have argued
that a characteristic of late modernity is the heightened importance of
reflexive construction of individual lives, resulting in a tendency
toward increasing heterogeneity in life trajectories. Beck has also
argued that the risks of poverty have become more
"democratized," increasingly reaching into the middle classes,
as well as "temporalized," and more likely to be short-term
(Beck 1992, 1999). These concerns are reflected in the life-course
literature as debates over whether life courses have indeed become more
individualized and heterogeneous, and whether temporal patterning has
become more or less strong (Henretta 2003).
Viewing welfare use from the perspective of the life course focuses
our interest on the various aspects of biography that affect benefit
take-up, the duration of welfare, and whether it is generally a short-
or long-term phenomenon, and the implications of welfare experiences for
later life events. Leisering and Leibfried's (1999) study of
welfare use in Germany is an excellent example of a life course approach
to state programs, investigating the reasons for welfare use, its
duration, and factors affecting exit from welfare. Thus far, there has
been little research on North American welfare systems that explicitly
uses this lens, with much of the existing research coming from
economics. However, that research can provide some important evidence
regarding the duration and dynamics of social assistance, and the
conditions that affect the timing of exit, and which can be interpreted
from a life course perspective.
PREVIOUS EVIDENCE ON SOCIAL ASSISTANCE DURATION AND THE
"WELFARE TRAP"
There has been some research on social assistance duration in
individual Canadian provinces and in the United States. Most previous
studies have incorporated various characteristics of individual
recipients, some measures of the local labor market, and some indicators
of the benefits available through welfare. Individual characteristics
are taken to affect one's ability to exit social assistance through
labor or marriage markets (Harris 1993; O'Neill et al. 1987).
Characteristics, such as education and previous work experience, as well
as marital status and the number and ages of children can be thought of
as reflecting human capital, but of course they are also reflections of
previously experienced transitions and trajectories in the domains of
work and education and family life (Cooke and Gazso 2009). As one would
expect, higher education and more work experience are consistently found
to increase the rate at which women leave social assistance, while the
presence of children, particularly young ones, make exit more difficult.
Whether a lone mother has been previously married is sometimes included,
and is generally found to increase the rate of exit from social
assistance, possibly through the availability of spousal benefits
(Harris 1993; O'Neill et al. 1987; Stewart and Dooley 1999).
O'Neill et al. (1987) also include an indicator of whether a woman
became a mother as a teen, but find no significant effect.
Aspects of the local labor market that are generally included in
these models are the expected wage rate or minimum wage, local
unemployment rates, and expected welfare benefits. Results have
confirmed that higher expected wages and lower unemployment increase the
pace of welfare exit (Harris 1993; O'Neill et al. 1987; Stewart and
Dooley 1999). One Quebec study found no clear effect of minimum wage
rates on exit times, however (Duclos et al. 1999).
In general, research has found that social assistance use is
characterized more by short-term than by long-term use. Using 1980s
administrative data from British Columbia and nonparametric models,
Barrett and Cragg (1998) find that most welfare spells were short, with
75 percent ending in six months, but that there was a high incidence of
return to welfare, with 50 percent of welfare leavers returning within a
year. They find that a quarter of all of these cases were single mothers
with children, and that lone mothers tended to leave welfare somewhat
more slowly. Importantly, the hazard function, representing the changing
"risk" of leaving social assistance, indicated that these BC
recipients were most likely to exit within six months of the beginning
of receipt, with the likelihood of exit declining thereafter. Duclos et
al. (1999) found that in Quebec 34 percent of new entrants to social
assistance between 1979 and 1993 left within six months, and estimated
the median duration in that province at about 22 months, suggesting that
the exit rate is initially high but drops off rather sharply. Their
estimate is similar to findings in the United States by Bane and
Ellwood, that half of AFDC spells in the 1980s were less than two years
(Bane and Ellwood 1994). Other evidence from the United States has also
found that a majority of lone mothers' spells are short, but that
there is an important number of long-duration recipients, and confirmed
that the probability of leaving decreases with time (O'Neill et al.
1987).
This question of "duration dependence," or how the risk
of leaving social assistance varies with rime, is an important one, as
it may provide evidence about one possible effect of social assistance
on people's subsequent life courses, and how a previous decision
affects the context of later ones. In particular, difficulty leaving
social assistance may be an "unintended consequence" of
receipt that is taken up to realize particular short-term goals
(Leisering and Leibfried 1999). This is sometimes referred to as the
question of the "welfare trap"--a decreasing ability to leave
social assistance. This effect may have several sources, including the
stigmatizing effects of long-term welfare receipt and inability to find
work, the degradation of human capital over time, psychological effects,
or some combination of these, although the exact cause is unclear
(Sandefur and Cook 1998).
This research was supported by a Social Sciences and Humanities
Research Council of Canada (SSHRCC) Doctoral Fellowship.
MARTIN COOKE University of Waterloo
Martin Cooke, Assistant Professor, Department of Sociology and
Department of Health Studies and Gerontology, University of Waterloo,
200 University Drive W, Waterloo, ON N2L 3G1. E-mail: cooke@uwa
terloo.ca.
Negative duration dependence is rather difficult to conclusively
identify and to sort out from spurious effects. Long-term cases can be
expected to accumulate over time, as those who are more able or willing
to leave social assistance do so. Some of this will be due to measurable
covariates, and can therefore be controlled in multivariate models with
appropriate data. However, as in demographic studies on migration and of
mortality, some of this effect may be due to unmeasured characteristics,
such as psychological states or preferences. In migration studies, it is
common to observe that some people are "movers" while others
are "stayers," and studies of mortality are often concerned
with controlling for the effects of unmeasured "frailty"
(Trussell and Richards 1985). As a result, the predicted probability of
an individual experiencing the event (migration, death, or exit from
social assistance) may spuriously appear to decrease with time because
the remaining sample becomes increasingly composed of those with an
unmeasured tendency to remain (Heckman and Singer 1984). In their study
of Canadian workers' compensation claims, Butler et al. (2001) find
that controlling for this "unobserved heterogeneity" has an
effect on parameter estimates, as well as on the observed duration
dependence. However, although most of the existing research on the
length of social assistance use acknowledges this problem, only one
Canadian study has attempted to control for it. Using 1990 to 1994
administrative Ontario data, this study confirmed the presence of
unobserved heterogeneity, but found that there remained a small negative
duration dependence effect (Stewart and Dooley 1999).
It should be noted that a declining hazard rate, even with
unobserved heterogeneity controlled, does not conclusively prove the
existence of a welfare trap. Using simulation models, Contini and Negri
(2007) have shown that a declining hazard rate can exist even in the
absence of welfare duration dependence, because it may be that time in
poverty or unemployment, rather than time on welfare, that leads to
declining exit rates. Because these trajectories are interrelated, and
data on the onset of poverty are generally limited, it is difficult to
isolate their independent effects. However, a declining exit rate is
certainly necessary for there to be a welfare trap. We discuss the
importance of this evidence further in our conclusions.
In this study, we use nationally representative survey data to
explicitly address questions related to the way the likelihood of
leaving social assistance depends on time. Using 1996 to 2001 data from
the Survey of Labour and Income Dynamics (SLID) and semiparametric and
parametric event history models, we investigate the duration of social
assistance receipt of lone mothers, the life course correlates of longer
receipt, and whether the risk of leaving social assistance does indeed
decline with time. Whereas previous Canadian studies have used
provincial data, these national SLID data widen our scope for inference,
while allowing us to include the effects of characteristics of the
previous life course and local labor market conditions.
DATA AND METHODS
The data used for this investigation are from Panel 2 (1996-2001)
of the SLID, a national prospective longitudinal survey of the finances
and incomes of Canadians (Statistics Canada 1994). (1) The SLID samples
include roughly 30,000 people from 16,000 households, who are selected
for inclusion in six-year overlapping panels. The survey includes
retrospective questions about marital and employment histories, and
collects prospective data on income, education and employment, benefit
use, and family and household changes. The SLID data also allow us to
identify lone mothers, defined for out purposes as the female heads of
lone parent census families aged 16 to 59, with at least one child at
home (Statistics Canada 2004). The SLID also collect data specifically
on monthly receipt of social assistance, providing up to 72 months of
observed social assistance receipt.
Independent variables include characteristics of the
individual's previous life course, including marital and household
status, age and number of children, education, and previous work
experience. Father's education is included as a rough proxy for the
social class of the respondent's birth family. As well, models
include provincial indicator variables as proxies for the different
rules and relative benefit levels associated with provincial systems. An
indicator variable controls for the presence of a work-limiting
disability. Last, the provincial minimum wage and regional benefit
levels are included as characteristics of local labor markets.
Provincial annual hourly minimum wage levels from 1995 to 2001 came from
Human Resources Development Canada (Human Resources Development Canada
2004), and were converted to constant 2001 dollars using the Consumer
Price Index provincial annual averages (Statistics Canada 2001, 2002a,
2002b). The average annual unemployment rates for Economic Regions
(Statistics Canada 1992) were taken from the Labour Force Survey data,
from the 2002 Labour Force Historical Review (Statistics Canada 2002a,
2002b). These variables are included in the multivariate models as
time-varying covariates, allowed to vary monthly.
We begin with nonparametric lifetable measures to describe the
differences in average social assistance duration for lone mothers and
other family types. Continuous-time event history models are used to
address the effects of the covariates on the length of time that an
individual spends receiving social assistance, and to investigate how
the likelihood of leaving depends on time previously spent on social
assistance. Semi-parametric models (Cox 1972) are used to estimate the
effects of the covariates on the hazards of leaving social assistance
for each of the 72 months of potential receipt. The Cox regression model
of the hazard rate for individual i can be expressed as
[h.sub.i](t) = [[lambda].sub.0](t)exp{[[beta].sub.1][x.sub.i1] +
... + [[beta].sub.k][x.sub.ik]} (1)
or the product of an unspecified baseline hazard function
[[lambda].sub.0](t) and an exponentiated set of k covariates (Allison
1995). Cox models are often favored because they do not assume a
specific distribution for the hazard function, and are subject only to
the major assumption that the shape of the function is the same for all
levels of the covariates. Violations of this assumption are generally
assessed and corrected by the inclusion of interactions of the
covariates with time (t).
Although the Cox models are flexible, and we can examine the shape
of the estimated baseline hazard function, they do not allow us to test
hypotheses about that shape. As well, the flexibility costs something in
efficiency of parameter estimation. In order to test specific hypotheses
about the shape of the hasard function, we estimate similar models in
parametric form using accelerated failure time (AFT) models. The general
form of the AFT model is
ln([t.sub.i]) = [[beta].sub.1][x.sub.i1] + ... +
[[beta].sub.k][x.sub.ik] + ln([[pi].sub.i]) (2)
where the distribution of ([[pi].sub.i]) takes a particular
distribution. These models are parametric models in the sense that both
covariates and the functional form of the dependence on time are
specified. The AFT metric more specifically addresses the issue of the
effects of time, and more easily allows predictions of failure times.
There may or may not be a firm theoretical reason to choose or
discount any particular functional form for the hazard. For our
purposes, we will fit models where the underlying process is assumed to
have an exponential distribution, a Weibull distribution, a log-normal
distribution, and a gamma distribution. These are important for our
tests of hypotheses about the effects of time. The exponential
distribution assumes a constant hazard rate, or that people experience
the event at the same rate over time. This is a special case of the
Weibull model, in which the hazard rate may change over time,
monotonically increasing or decreasing. The log-normal distribution is
nonmonotonic, allowing the hazard to increase and then decrease, as
would be the case if people were more likely to leave social assistance
in the first few months of receipt, followed by a decline. The
log-normal and the Weibull distribution are special cases of the gamma
distribution, and they can be compared using a likelihood ratio
chi-square test (Heckman and Walker 1987). Last, in order to test for
spurious duration dependence, or "unobserved heterogeneity,"
the AFT models were re-estimated with gamma-distributed,
individual-specific random terms, using the facility provided in Stata
v.8. This procedure attempts to model the heterogeneity in the sample by
creating a mixture model in which the heterogeneity is assumed to takes
some flexible distribution. By testing the significance of the
heterogeneity term's parameters, we can judge whether duration
dependence is likely to be spuriously produced by the accumulation of
long-duration cases in the sample, or whether it represents the real
risk of social assistance exit over time (Cleves et al. 2004)
There are some problems with the SLID data. For one thing, the
short duration of the panel means that censoring, the nonobservation of
the beginning or the end of a social assistance spell, may be heavy. To
the extent possible, we have addressed this through our choice of
methods, particularly the use of proportional hazards models and partial
likelihood estimation. The monthly social assistance data also forces
form choices about what constitutes an "end" to social
assistance. Ideally, the observation period would be long enough and the
sample large enough to observe all social assistance spells, and to
model the duration of first, second, and subsequent spells separately,
as well as the probability of subsequent spells and the final exit from
welfare. Instead, we model the duration of the first observed spell,
defined to have ended after two months of nonreceipt, in order to
disregard "administrative" endings (Stewart and Dooley 1999).
Also, we model the total number of observed months on social assistance
in order to observe whether there are different effects due to the way
in which social assistance receipt is defined (Dahl and Lorentzen 2003).
The time variable in the total duration models is the total number of
months of observed receipt, and these models therefore do not include
the monthly measures of the local labor market conditions.
There have been problems with the reporting of months of social
assistance usage in the SLID, compared with administrative data, and the
"heaping" beginnings and exits from social assistance at the
beginning and end of the calendar year. Kapsalis (2001) finds that the
underreporting of approximately 15 percent might be ignorable, and also
that those terminating in December and beginning in January were not
significantly different than other recipients, on a number of
characteristics.
The final samples for the event history analyses are presented in
Table 1 for all family major income earners in census families,
including lone mothers. After deleting cases who reported receiving
social assistance, but not the months of receipt, there were 1,940 major
income earners who received social assistance at least one month during
the 1996 to 2001 period, 635 of whom were lone mothers at the time of
receipt. Of those, 577 had complete records and are included in Table 2.
RESULTS
Following our general analytic strategy, we first present
nonparametric estimates of the length of social assistance, comparing
lone mothers with the heads of other household types. We then present
semiparametric multivariate models for all household types, and for lone
mothers separately. Last, we present the results of the best-fitting AFT
models and interpret the results in terms of the predicted length of
social assistance receipt.
Length of Social Assistance
Long-term social assistance receipt was hot common in the SLID
data. Without adjusting for censoring, 10 percent of the women in the
sample who experienced lone motherhood and social assistance between
1996 and 2001 received benefits for six months or less and 20 percent
experienced a total observed duration that was between six months and
one year. Fifty-seven percent were observed to spend two years or less
on social assistance, and 15 percent received benefits for two to three
years. Eighteen percent received benefits for four years or more, 8
percent for five years or longer, and 5 percent of lone mothers who
received social assistance did so continuously for the entire six years
of the panel.
Table 3 presents the nonparametric median time to exit, for the
first, second, and third observed spells, and for the total observed
duration. The duration of the first observed spells and the total
observed duration were significantly longer for lone mothers than for
other family types, 27 months, compared with 23.4 months. Second and
third spells were also more likely to be observed for lone mothers,
despite the fact that the longer first spells left somewhat less time in
which to observe subsequent spells.
Predictors of the Duration of Social Assistance Receipt
Lone mothers also exited social assistance more slowly than major
income earners in other families, once we controlled for education, age,
work experience, fathers' education, disability, province, and
local labor market conditions. Table 4 presents the semiparametric
models of the total observed social assistance duration and the first
observed spell for al family types. Interactions with functions of time
were included to correct for violations of the proportionality
assumption. Lone mothers had significantly longer first observed spells
and total observed social assistance duration than couples with
children, and in both models exited about 33 percent more slowly (Table
4).
These models also tell us something about the aspects of the life
course, besides being a lone mother, that influence the duration of
social assistance. As shown in Table 4, controlling for the other
variables, age significantly decreased the rate of exit from social
assistance, although the effect of each additional year was only slight
(HR = .99). Surprisingly, there were few significant independent effects
of education. Having less than high school decreased the rate of exit
from the first observed spell by about 15 percent. Previous work
experience had a greater effect, with those with more years exiting
earlier. Fathers' education had no significant independent effect
on the rate of exit of the first spell, or on the total observed
duration. As expected, a higher regional unemployment rate significantly
reduced the pace of exit, while higher minimum wages significantly
increased exit rates.
Table 5 presents the results of models of the total observed
duration and the first observed spell of social assistance for women who
were lone mothers when they began that spell. These models include some
aspects of the life course specific to lone mothers, such as the age at
which a woman had her first child and the number of preschool children
living with her. Among women who had been married, the age of first
marriage was included in some models, but was collinear with age at
first birth. Included separately, it did not improve the model fit more
than the marital status variable, and was thus excluded. These models
did not seriously violate the proportionality assumption, so no
interactions with time were included.
Lone mothers who had never been married received social assistance
significantly longer than those who were divorced, separated, or
widowed. Never-married lone mothers exited at a rate that was 30 percent
slower than those who had been married and were divorced or widowed at
the beginning of their first observed spell. Older women tended to leave
social assistance more slowly than younger women, but age had no
significant effect on the duration of the first observed spell alone.
Models with age as a set of dummy variables offered no significant
improvement in model fit judging by a likelihood ratio chi-square test,
and showed no evidence of a nonlinear age relationship (not shown).
Although lone mothers who were older when they had their first children
were more likely than their younger counterparts to receive social
assistance in 2001, the age at first birth had no significant effect on
the length of time spent on social assistance. Furthermore, age at first
marriage did not significantly affect the length of time respondents
spent receiving social assistance (models not shown).
Similar to the models for all family types, education had no
significant effect on the length of time a lone mother received social
assistance. This may be because of the coarse categorization of
education due to small numbers of lone mothers with university degrees
who experienced social assistance, or to lower statistical power because
of smaller sample size. However, it also may be that lone mothers with
higher education were less likely to experience social assistance, as
Charette and Meng (1994) found, but that once on social assistance,
those with higher education did not leave social assistance
significantly faster. As with earlier models, the level of education of
a woman's father, a proxy for resources available from family, also
had no significant independent effect on the rate at which she left
social assistance.
Lone mothers who had more years of work experience may leave social
assistance more quickly, controlling for chronological age as well as
the other variables, but the evidence from the models is not clear.
Those with 10 to 19 years of experience had a predicted hazard of
leaving social assistance that was 37 percent higher than those with no
work experience, but this effect was significant only for total
duration, and not for the first observed spell. This could indicate that
those with more work experience were less likely to experience more than
one spell. None of the other work experience indicators were significant
at the .05 level, however (Table 5).
Women with preschool children at home did not experience
significantly longer periods of social assistance receipt, controlling
for the other variables. Recall that the number of preschool children
living with a lone mother also did not significantly affect the
probability that she received social assistance. Again, alternative
models that instead included the total number of children of all ages
also found that variable to be insignificant, controlling for other
variables in the model.
[FIGURE 1 OMITTED]
[FIGURE 2 OMITTED]
Provincial effects were somewhat different for lone mothers than
for major income earners of all family types combined. As with the model
including all census family major income earners, higher unemployment
was associated with longer durations on social assistance, while a
higher minimum wage tended to reduce the length of the first observed
spell.
The Shape of the Hazard Function: Changing Exit Risks Over Time
To address our major interest of the way that social assistance
exits depend on time, we first examine the smoothed estimated baseline
hazard from the Cox regression models. Figure 1 presents a plot of that
function for the model presented in Table 5, roughly interpreted as the
changing instantaneous probability of leaving the first spell of social
assistance. The general shape of the hazard indicates a rising risk of
exit, peaking just before the 18th month of receipt, and declining
rather sharply afterward. Separate curves are presented for lone mothers
who were separated, divorced or widowed, or single and never married, at
the beginning of the first observed spell. The significantly lower risk
of exit, and therefore the higher duration of social assistance, is
visible for women who had never been married.
[FIGURE 3 OMITTED]
As described above, the proportional hazards model does not let us
test specific hypotheses about the shape of the hazard function. By
estimating exponential, Weibull, log-normal, and gamma AFT models, we
were able to determine which best fit the data. Likelihood ratio tests
show that the gamma is not an improvement over either the Weibull or the
log-normal models, leading to the conclusion that if the hazard does
increase and then decrease, there is only a single peak. The parameter
estimates and standard errors of the models changed only very slightly
after inclusion of the frailty term, and so we conclude that the
duration dependence does not appear to unobserved heterogeneity and the
accumulation of long-term recipients in the sample. In other words, we
find that leaving social assistance becomes less likely with additional
time in receipt, both for lone mothers and for the heads of other family
types (models not shown).
Model Results
The results of the log-normal model are not presented, as the
direction and significance of the covariates in the log-normal AFT model
were similar to those in the earlier Cox models. Lone mothers who had
never been married tended to exit the first observed spell of social
assistance more slowly than those who had been married and were lone
mothers after divorce or the death of a spouse. Figure 2 shows the
estimated hazard of leaving the first spell for lone mothers who were
divorced or widowed and those who were never married. The risk of
leaving social assistance rises during the first few months and then
declines, but more steeply for those who had been married.
This is also shown in the survival curve for the same model, which
indicates the proportion of lone mothers who can be expected to remain
on social assistance, by months already spent in receipt. This shows
that more lone mothers who were never married can be expected to stay on
social assistance for the entire 72-month period (Fig. 3). The predicted
median length of the first observed spell for a lone mother who was
never married, who was 33 years old at the start of the spell, and who
had average or reference values on the other covariates was 28 months.
This is compared with a predicted median duration of 20 months for a
divorced or widowed lone mother.
CONCLUSIONS
We have proposed a life course framework for the interpretation of
these results, for several reasons. It helps us to incorporate
scholarship that focuses on the role of the welfare state in reproducing
class and gender relations and in creating expectations about the timing
of education, work, and retirement. It also directs attention to the
choices made by individuals and families within particular social
contexts (Marshall and Mueller 2003). Perhaps most importantly, the life
course offers a dynamic approach to social assistance receipt and
reminds us that decisions and events experienced in the past affect the
resources with which future challenges are faced. The Canadian federal
Policy Research Initiative (PRI) has proposed a framework for policy
analysis that would examine how these resources are accumulated and
depleted across the life course and the times and situations in which
social, financial, human capital, or other resources are most likely to
be needed. The goal of life-course-sensitive policy would be to develop
policies that provide the appropriate kinds of support at appropriate
times, and to help people meet their various goals (PRI 2004).
Viewed through a life course lens, the SLID data tell us something
about how life course characteristics are reflected in social assistance
receipt. Unsurprisingly, family heads with less education and work
experience tend to experience longer periods of receipt. Gender is an
important aspect of social assistance policy in the sense that women, as
lone mothers, are more likely to be recipients, but lone mothers also
receive benefits for longer, although the average durations presented
above do not seem to be consistent with a popular view of long-term
"welfare dependency." Education had much less impact on the
predicted durations for lone mothers than other household heads, and
this was also true for work experience, although previous years at work
did have some effect for lone mothers. What seems to matter more, in
terms of the duration of assistance, is a lone mother's marital
history. Lone mothers who had been married might be more likely to have
spousal or parental benefits. As well, living with other adults, such as
her parents, might reflect a source of social support, possibly
including help with child care. This suggests that receipt may not be
best seen as a simple choice between welfare and work, and that social
support and other resources may be more important than human capital for
reducing time on social assistance.
The models presented here explicitly investigate how the likelihood
of leaving social assistance depends on time, and provide evidence that
that the longer one receives social assistance, the less likely one may
be to leave it. By using frailty models to test for unobserved
heterogeneity, we have been able to show that this is not the spurious
result of unobserved characteristics related to long-term receipt, but
is likely to be a real reflection of how lone mothers' likelihood
of leaving welfare changes with time spent receiving benefits. For lone
mothers, predicted exit probabilities in our best models were much
higher in the first few months of receipt, and declined afterward. In
other words, the longer one receives benefits, the less likely one is to
leave social assistance.
This strongly suggests the presence of a "welfare trap"
effect in which alternatives to welfare become increasingly difficult to
pursue. As described above, the declining hazard rate does not
conclusively demonstrate that it is social assistance itself that makes
escaping it more difficult with time. Indeed, these results could be
reflections of trajectories of poverty or unemployment that include
spells of social assistance receipt, while social assistance receipt
itself is not the cause of the "welfare trap." Loss of
self-esteem or human capital may result from time in poverty or out of
the workforce, making it harder to leave social assistance once receipt
has begun. Social stigma affecting the likelihood of finding a job, and
the psychological resources to deal with a job search, might be caused
by poverty as much as by welfare receipt (Contini and Negri 2007).
However, there is reason to think that welfare itself might have an
inhibiting effect on exits, and this might be the result of the types of
welfare policy in place. A system that requires that most other sources
of support be exhausted before turning to welfare might not actually
support transitions to independence, but instead make those transitions
more difficult. The current emphasis on transitions into the first
available job (Gorlick and Brethour 1998), including jobs that lone
mothers might reasonably know are unlikely to provide long-term
"independence" and income enough to support themselves and
their children, might lead some women to resign themselves to benefit
receipt. This may be particularly the case in the absence of a system of
low-cost childcare. Moreover, the highly stigmatized nature of social
assistance in Canada, as in the other Anglo-American welfare states,
might indeed have effects on exits that are independent of the already
stigmatizing effects of poverty and unemployment.
From this perspective, increased difficulty leaving social
assistance with time could be thought of as an "unintended
outcome" of the decision to receive benefits (Leisering and
Leibfried 1999). People may begin receipt in order to realize a
particular short-term goal or to deal with a crisis, such as leaving an
abusive relationship, but then find it increasingly difficult to exit,
particularly through work. This may be the case in general, but even
more for lone mothers, whose resources may be more limited and needs
greater than those of couples or unattached individuals.
If it is true that education and work experience do not help women
leave social assistance faster, or at least do not have strong effects,
the recent policies of mandatory work effort and training programs would
be somewhat misguided. It may also be that the means and income tests
for most programs, which make it necessary to nearly exhaust all savings
and liquidate assets before turning to welfare, eliminate at least one
advantage that women with previous work histories might have
had--accumulated wealth or personal savings. At the same time, the human
capital implied by work experience and education may be difficult to
exploit because of the requirement that any work should be flexible
enough to allow the combination of work and care giving. Although these
models are not definitive, once a woman is a lone mother on social
assistance, her previous education or work experience might not be
translated into assets that can help her leave welfare.
A life course view suggests policies that support people in making
transitions and avoiding "traps" of long-term receipt and
social exclusion. However, in order to better understand the role of
these policies in shaping trajectories of poverty, we suggest two
directions for future research. First, social assistance trajectories
should be examined and related to poverty trajectories more generally,
including income and employment trajectories and the roles of various
welfare state programs, such as EI, social assistance, pensions, and
disability benefits, within individual lives. This may help to
conclusively identify the existence of a "welfare trap" or
other negative effects of particular programs, and hopefully lead to
policy suggestions for decreasing these effects. Unfortunately, this
research would likely depend on the existence of long-term socioeconomic
panel data like those that exist in other countries but are not yet
available in Canada.
Second, investigation of the resources that help recipients leave
social assistance, including social support, various financial
resources, health, and others, would improve understanding of the
process of exiting welfare, a research topic that is quite undeveloped.
This may also require better panel data than are currently available
from sources like the SLID, but would also benefit from a qualitative
understanding of the strategies and resources used to exit social
assistance from the perspective of recipients themselves. Having a
better understanding of the goals and constraints in the lives of lone
mothers on social assistance may help generate policies that provide the
types of support that are most helpful.
Finally, the changing economic situation may affect the way we
think about social assistance and the life course. The data analyzed
here are from years during which Canada had substantial economic growth,
relatively low unemployment, and declining social assistance rates.
Recently, rising unemployment suggests that social assistance use will
again increase among lone mothers and other family types. This could be
more the case than during the recession of the 1980s, when eligibility
for EI was easier (Campeau 2005). The question of whether the risk of
unemployment and poverty has truly been "democratized" will be
only visible in hindsight but the policy focus on the reduction of
"dependency" may become less tenable as the availability of
work becomes further restricted, and if the composition of the welfare
caseload includes more men and two-parent families, and reaches higher
into the middle class. In this paper, like other recent analysts, we
focus on the importance of social assistance in individual biographies
and largely as a response to events, such as divorce or the birth of a
child. Significant growth in welfare rates may turn the attention of
analysts to the role of social assistance in protecting against broadly
felt externally generated economic risks. It may also generate political
pressure for a return to higher benefit levels, although that remains to
be seen.
References
Allison, P.D. 1995. Survival Analysis in SAS. Carey, NC: SAS
Institute.
Bane, M.J. and D.T. Ellwood. 1994. Welfare Realities: From Rhetoric
to Reform. Cambridge, MA: Harvard University Press.
Barrett, G.F. and M.I. Cragg. 1998. "An Unteld Story: The
Characteristics of Welfare Use in British Columbia." The Canadian
Journal of Economics 31:165-88.
Bashevkin, S. 2002. Welfare Hot Buttons: Women, Work, and Social
Policy Reform. Toronto: University of Toronto Press.
Beck, U. 1992. Risk Society: Towards a New Modernity. London: Sage.
Beck, U. 1999. World Risk Society. Cambridge, UK: Polity Press.
Beck, U., A. Giddens, and S. Lash. 1994. Reflexive Modernization:
Politics, Tradition and Aesthetics in the Modern Social Order. Stanford:
Stanford University Press.
Boychuk, G. 2006. "Slouching Toward the Bottom: Social
Assistance Provision in Canada, 1980-2000 Racing to the Bottom?"
Pp. 157-92 in Provincial Interdependence in the Canadian Federation,
edited by K. Harrison. Vancouver: UBC Press.
Butler, R.J., M.L. Baldwin, and W.G. Johnson. 2001. "The
Effects of Worker Heterogeneity on Duration Dependence: Low-Back Claims
in Workers Compensation." The Review of Economics and Statistics
83:708-16.
Campeau, G. 2005. From UI to EI: Waging War on the Welfare State.
Vancouver: UBC Press.
Charette, M. and R. Meng. 1994. "The Determinants of Welfare
Participation of Female Heads of Household in Canada." Canadian
Journal of Economics 27:290-306.
Cleves, M.A., W.W. Gould, and R.G. Gutierrez. 2004. An Introduction
to Survival Analysis Using Stata, Rev. ed. College Station, TX: Stata
Press.
Contini, D. and N. Negri. 2007. "Would Declining Exit Rates
from Welfare Provide Evidence of Welfare Dependence in Homogenous
Environments?" European Sociological Review 23:21-33.
Cooke, M. and A. Gazso. 2009. "Taking a Life Course
Perspective on Social Assistance Use in Canada: A Different
Approach." Canadian Journal of Sociology 34(2):347-70.
Cox, D.R. 1972. "Regression Models and Lire-Tables (with
Discussion)." Journal of the Royal Statistical Society Series B
4(2):505-33.
Dahl, E. and T. Lorentzen. 2003. "Dynamics of Social
Assistance: The Norweigian Experience in Comparative Perspective."
International Journal of Social Welfare 12(4):289-301.
Daly, M. and K. Rake. 2003. Gender and the Welfare State.
Cambridge, UK: Polity Press.
Duclos, J.-Y., B. Fortin, G. Lacroix, and H. Roberge. 1999.
"The Dynamics of Welfare Participation in Quebec." Pp. 191-224
in Women and Work, edited by R.P. Chayykowski and L.M. Powell. Kingston:
McGill-Queen's University Press.
Edin, K. and L. Lein. 1996. "Work, Welfare, and Single
Mothers' Economic Survival Strategies." American Sociological
Review 61:253-66.
Elder, G.H. Jr. 1994. "Time, Human Agency, and Social Change:
Perspectives on the Life Course." Social Psychological Quarterly
57:4-15.
Esping-Andersen, G. 1999. Social Foundations of Postindustrial
Economies. Oxford, NY: Oxford University Press.
Frenette, M. and G. Picot. 2003. "Life After Welfare: The
Economic Well Being of Welfare Leavers in Canada During the 1990s."
Research Paper Series, Statistics Canada, Ottawa.
Gorlick, C. and G. Brethour. 1998. Welfare-to-Work Programs: A
National Inventory. Ottawa: Canadian Council on Social Development.
Harris, K.M. 1993. "Work and Welfare Among Single Mothers in
Poverty." American Journal of Sociology 99:317-52.
Heckman, J. and B. Singer. 1984. "A Method for Minimizing the
Impact of Distributional Assumptions in Econometric Models for Duration
Data." Econometrica 52:271-320.
Heckman, J.J. and J.R. Walker. 1987. "Using Goodness of Fit
and Other Criteria to Choose among Competing Duration Models: A Case
Study of Hutterite Data." Sociological Methodology 17:247-307.
Henretta, J. 2002. "The Life-Course Perspective on Work and
Retirement." Pp. 85-105 in Invitation to the Life Course: Toward
New Understandings of Later Life, edited by R.A. Settersten Jr.
Amityville, NY: Baywood.
Human Resources Development Canada. 2004. Hourly Minimum Wages in
CANADA for Adult Workers 1995-2004. Ottawa: HRDC.
Kapsalis, C. 2001. "An Assessment of EI and SA Reporting in
SLID." Analytical Studies Branch Research Paper Series No. 166.
Ottawa: Statistics Canada.
Kohli, M. 1986. "The World We Forgot: A Historical Review of
the Life Course. Later Life." Pp. 271-303 in The Social Psychology
of Ageing, edited by V.W. Marshall. Beverly Hills: Sage.
Leisering, L. and S. Leibfried. 1999. Time and Poverty in Western
Welfare States: United Germany in Perspective. Cambridge, UK: Cambridge
University Press.
Little, M.H. 1998. "No Car, No Radio, No Liquor Permit":
The Moral Regulation of Single Mothers in Ontario. Toronto: Oxford
University Press.
Marshall, V.W. and M.M. Mueller. 2003. "Theoretical Roots of
the Life-Course Perspective." Pp. 3-32 in Social Dynamics of the
Life Course: Transitions, Institutions, and Interrelations, edited by
W.R. Heinz and V.W. Marshall. New York: Aldine de Gruyter.
McMullin, J.A., L. Davies, and G. Cassidy. 2002. "Welfare
Reform in Ontario: Tough Times in Mothers' Lives." Canadian
Public Policy xxviii:297-314.
Mink, G. 2001. "Violating Women: Rights Abuses in the Welfare
Police State." Special Edition of the Annals of the American
Academy of Political and Social Science 577:79-93.
Myles, J. and P. Pierson. 1997. "Friedman's Revenge: The
Reform of 'Liberal' Welfare States in Canada and the United
States." Politics and Society 25:443-72.
O'Neill, J.A., L.J. Bassi, and D.A. Wolf. 1987. "The
Duration of Welfare Spells." The Review of Economics and Statistics
69:241-48.
O'Rand, A.M. 1996. "The Precious and the Precocious:
Understanding Cumulative Disadvantage and Cumulative Advantage Over the
Life Course." The Gerontologist 3:230-38.
Picot, G., R. Morrisette, and J. Myles. 2003. "Low-Income
Intensity During the 1990s: The Role of Economic Growth, Employment
Earnings and Social Transfers." Analytical Studies Branch Research
Paper Series, Statistics Canada, Ottawa.
Pierson, P. 1994. Dismantling the Welfare State? Reagan, Thatcher
and the Politics of Retrenchment. Cambridge, UK: Cambridge University
Press.
Piven, F.F. and R. Cloward. 1993. Regulating the Poor: The
Functions of Public Welfare, Updated ed. New York: Vintage Books.
Policy Research Initiative (PRI). 2004. A Life-Course Approach to
Social Policy Analysis: A Proposed Framework. Population Aging and
Life-Course Flexibility. Ottawa: Policy Research Initiative.
Sandefur, G.D. and S.T. Cook. 1998. "Permanent Exits from
Public Assistance: The Impact of Duration, Family, and Work."
Social Forces 77:763-87.
Settersten, R.A. Jr. 2003. "Propositions and Controversies in
Life-Course Scholarship." Pp. 15-45 in Invitation to the Life
Course: Towards New Understandings of Later Life, edited by R.A.
Settersten Jr. Amityville: Baywood.
Statistics Canada. 1992. Census Dictionary 1991. Ottawa: Industry
Canada.
Statistics Canada. 1994. Survey of Labour and Income Dynamics
User's Guide Wave 2, 1994. Ottawa: Minister of Industry.
Statistics Canada. 2001. The Consumer Price Index January, 2001.
Consumer Price Index. Ottawa: Statistics Canada.
Statistics Canada. 2002a. The Consumer Price Index January, 2002.
The Consumer Price Index. Ottawa: Statistics Canada.
Statistics Canada. 2002b. "Labour Force Estimates by Economic
Region, Annual Average." 2002 Labour Force Historical Review.
Ottawa.
Statistics Canada. 2004. "Definitions of Concepts and
Variables-Census Family." Retrieved June 15, 2004,
(http://www.statcan.ca/english/concepts/definitions/eco-family.htm).
Stewart, J. and M.D. Dooley. 1999. "The Duration of Spells on
Welfare and Off Welfare among Lone Mothers in Ontario." Canadian
Public Policy xxv(Suppl 1):s47-s72.
Trussell, J. and T. Richards. 1985. "Correcting for Unmeasured
Heterogeneity in Hazard Models Using the Heckman--Singer
Procedure." Sociological Methodology 15:242-76.
Wilthagen, T. 2007. Flexicurity Pathways: Expert Group on
Flexicurity Interim Report. Brussels: European Commission.
(1) These data were analyzed at the Statistics Canada South Western
Ontario Research Data Centre, using the SLID toaster file. The analysis
used data from Statistics Canada. The results and views expressed here
do net represent the views of Statistics Canada.
Table 1
Sample Description for Event History Models: Major Income
Earners Receiving Social Assistance
Number %
Census family status at start of first spell
Married/common-law w/children 333 17.2
Married/common-law no children 115 5.9
Unattached individual 806 41.6
Female lone parent 635 32.7
Other 51 2.6
Education at start of first spell
Less than high school 771 39.7
High school 561 30.8
Nonuniversity postsecondary 427 22.0
University degree 61 3.14
Missing/don't know 120 6.19
Years full-time work experience
No work experience 515 26.6
1-9 years 622 32.1
10-19 years 333 17.2
> 20 years 236 12.2
Don't know/refused 234 12.1
Father's education
Less than high school 1,189 61.3
High school 236 12.2
Degree or certificate 165 7.0
Don't know 350 18.0
Disability 669 36.0
Province (at start of first spell)
Newfoundland/Labrador 116 6.0
Prince Edward Island 46 2.4
Nova Scotia 127 6.6
New Brunswick 138 7.1
Quebec 496 25.6
Ontario 553 28.5
Manitoba 89 4.6
Saskatchewan 105 5.4
Alberta 97 5.0
British Columbia 173 8.9
Total 1,940 100
Mean SD
Age at start of spell 36.42 10.90
Monthly provincial SA benefits ($) 847.64 346.40
Regional unemployment rate 9.87 4.16
Minimum wage ($/hour) 6.00 .73
Source: SLID Panel 2.
SLID, Survey of Labour and Income Dynamics.
Table 2
Sample Description for Event History Models: Female Lone
Parents Receiving Social Assistance
Number %
Marital status
Divorced or widowed 177 30.7
Separated 137 23.7
Single, never married 263 45.6
Number of preschool children
None 326 56.5
One 193 33.5
More than one 58 10.1
Education
Less than high school 190 32.9
High school 214 37.1
Degree or certificate 173 30.0
Work experience
No work experience 171 29.5
1-10 years 231 40.0
10-19 years 112 19.3
> 20 years 33 5.7
Don't know or refused 30 5.2
Father's education
Less than high school 371 64.3
High school 85 14.7
Degree or certificate 40 6.9
Don't know 72 12.5
Disability 125 21.7
Province
Newfoundland/Labrador 25 4.3
Prince Edward Island 18 3.1
Nova Scotia 45 7.8
New Brunswick 49 8.5
Quebec 105 18.2
Ontario 186 32.2
Manitoba 35 6.0
Saskatchewan 35 6.0
Alberta 22 3.8
British Columbia 59 10.2
Total 577 100
Mean SD
Age at start 33.35 8.76
Age at first birth 22.26 4.46
Monthly SA benefits 1,078.16 175.32
Regional unemployment rate 9.51 3.76
Minimum wage ($/hour) 5.98 8.76
SLID, Survey of Labour and Income Dynamics.
Source: SLID Panel 2.
Table 3
Life Table Estimates of Median Time to Failure, First, Second,
and Third Observed Spells and Total Observed Duration of Social
Assistance Receipt by Major Income Earners, 1996-2001
Median
months SE of %
to exit median N % Censored
All census family major income earners
First spell 18.73 .4103 2,401 100 25.8
Second spell 13.60 .6810 681 28.3 30.3
Third spell 9.30 .8270 132 5.5 31.8
Total duration 23.42 .7362 2,401 -- 25.8
Female lone parents
First spell 21.92 1.128 686 100 23.3
Second spell 12.46 1.366 222 32.4 26.6
Third spell 9.71 1.361 52 7.8 21.2
Total duration 27.12 1.047 686 -- 23.3
Source: SLID, Panel 2.
SLID, Survey of Labour and Income Dynamics.
Table 4
Cog Proportional Hazards Models of Social Assistance Durations
by Census Family Major Income Earners, 1996-2001
Total observed duration
Hazard Pr > [chi
Coeffecient ratio SE square]
Census family status (married/C-L, with children = ref)
Married/C-L, no child .188 1.207 .130 .148
Unattached individual 4.097 60.159 .190 .000
Female lone parent -.412 .662 .080 .000
Other family type -.078 .925 .170 .647
Age -.012 .988 .003 .001
Education (high school = ref)
Less than high school -.084 .919 .068 .219
Nonuniversity .122 1.130 .075 .104
University .188 1.207 .153 .219
D.K./refused .048 1.049 .165 .771
Years Work Experience (1-9 years = ref)
No work experience -.239 .787 .076 .002
10-19 years .261 1.298 .081 .001
> 20 years .197 1.218 .106 .062
D/K or refused .103 1.109 .119 .386
Father's education (high school = ref)
Less than high school -.086 .918 .082 .294
Degree or certificate .097 1.101 .120 .421
Don't know .003 1.003 .103 .980
Disability Status
(no = ref) .949 2.584 .104 .000
Provinces (Ontario = ref)
Newfoundland/PEI -.240 .786 .111 .031
Nova Scotia -.228 .796 .118 .053
New Brunswick -.439 .644 .120 .000
Quebec -.256 .774 .075 .001
Manitoba -.098 .907 .135 .471
Saskatchewan .135 1.144 .123 .272
Alberta .241 1.273 .131 .065
British Columbia .758 2.134 .142 .000
Regional unemployment rate -- -- -- --
Minimum wage -- -- -- --
Unattached individual x
ln(time) -1.341 .262 .059 .000
Disability status x time -.040 .961 .003 .000
British Columbia x time -.035 .965 .006 .000
N 1,940
% Censored 27.0
- 2LL 18,089.100
First observed spell
Hazard Pr > [chi
Coefficient ratio SE square]
Census family status (married/C-L, with children = ref)
Married/C-L, no child .120 1.127 .122 .328
Unattached individual 2.905 18.268 .160 .000
Female lone parent -.387 .679 .075 .000
Other family type -.229 .795 .163 .159
Age -.012 .988 .003 .000
Education (high school = ref)
Less than high school -.155 .557 .064 .015
Nonuniversity .040 1.040 .071 .577
University .279 1.322 .143 .051
D.K./refused -.228 .796 .153 .137
Years Work Experience (1-9 years = ref)
No work experience -.199 .819 .070 .004
10-19 years .161 1.175 .076 .034
> 20 years .281 1.324 .097 .004
D/K or refused .170 1.185 .113 .133
Father's education (high school = ref)
Less than high school -.020 .980 .077 .794
Degree or certificate .061 1.063 .114 .594
Don't know .079 1.083 .097 .414
Disability Status
(no = ref) -.110 .895 .096 .248
Provinces (Ontario = ref)
Newfoundland/PEI .848 2.335 .202 .000
Nova Scotia -.106 .899 .184 .563
New Brunswick .807 2.240 .266 .002
Quebec -.570 .566 .204 .005
Manitoba .412 1.509 .204 .043
Saskatchewan -.204 .815 .157 .194
Alberta .932 2.539 .192 .000
British Columbia .427 1.532 .127 .001
Regional unemployment rate -.037 .964 .010 .000
Minimum wage .543 1.721 .093 .000
Unattached individual x
ln(time) -1.115 .328 .054 .000
Disability status x time -.010 .990 .003 .003
British Columbia x time -.027 .974 .005 .000
N 1,940
% Censored 16.1
- 2LL 21,444.078
Table 5
Cog Proportional Hazards Model of Social Assistance Durations
for Female Lone Parents, 1996-2001
Total observed duration
Hazard Pr > [chi
Coefficient ratio SE (B) square]
Marital Status (Divorced/Vidowed = ref)
Separated -.019 .981 .138 .888
Single, never married -.368 .692 .140 .009
Age at start of spell -.031 .969 .011 .003
Age at first birth .005 1.005 .014 .751
Education (high school = ref)
Less than high school -.074 .929 .126 .555
Degree or certificate .188 1.207 .122 .123
Years work experience (1-9 years = ref)
No work experience -.255 .775 .132 .053
10-19 years .314 1.370 .139 .024
> 20 years -.087 .916 .264 .741
D/K or refused .593 1.810 .243 .015
Father's education (high school = ref)
Less than high school -.170 .844 .141 .229
Degree or certificate .381 1.463 .210 .069
Don't know -.088 .916 .191 .645
Number of preschool children (0 = ref)
One -.268 .765 .148 .070
Two or more -.353 .702 .191 .064
Disability status
(no = ref) -.382 .683 .128 .003
Provinces (Ontario = ref)
Newfoundland/PEI -.064 .938 .204 .752
Nova Scotia -.259 .772 .206 .208
New Brunswick -.483 .617 .204 .018
Quebec -.301 .740 .151 .046
Manitoba .042 1.043 .215 .845
Saskatchewan .586 1.798 .209 .005
Alberta .724 2.064 .254 .004
British Columbia .129 1.135 .179 .472
Lives with other adults .115 1.122 .169 .495
Regional unemployment rate -- -- -- --
Minimum wage -- -- -- --
N 577
% Censored 24.09
- 2LL 3,398.559
First observed spell
Hazard Pr > [chi
Coefficient ratio SE (B) square]
Marital Status (Divorced/Vidowed = ref)
Separated -.113 .893 .132 .392
Single, never married -.361 .697 .140 .010
Age at start of spell -.015 .985 .009 .103
Age at first birth -.009 .991 .013 .475
Education (high school = ref)
Less than high school -.189 .827 .118 .109
Degree or certificate .118 1.125 .112 .295
Years work experience (1-9 years = ref)
No work experience -.156 .856 .123 .206
10-19 years .125 1.133 .132 .343
> 20 years .239 1.269 .228 .296
D/K or refused .245 1.277 .239 .305
Father's education (high school = ref)
Less than high school -.170 .844 .129 .188
Degree or certificate .111 1.118 .207 .591
Don't know .017 1.017 .171 .920
Number of preschool children (0 = ref)
One -.091 .913 .132 .491
Two or more -.006 .994 .173 .970
Disability status
(no = ref) -.280 .756 .119 .019
Provinces (Ontario = ref)
Newfoundland/PEI 1.227 3.409 .425 .004
Nova Scotia -.144 .566 .327 .660
New Brunswick .695 2.004 .517 .179
Quebec -.475 .622 .378 .209
Manitoba .432 1.541 .382 .257
Saskatchewan -.219 .803 .293 .455
Alberta 1.232 3.428 .356 .001
British Columbia .057 1.059 .170 .737
Lives with other adults .357 1.429 .155 .021
Regional unemployment rate -.066 .936 .022 .003
Minimum wage .503 1.653 .188 .008
N 577
% Censored 13.52
- 2LL 5,526.5242
